Google just rolled out searchable transcripts for videos stored in Google Drive, rolling out to all Workspace users by March 26. Transcripts appear in a YouTube-like sidebar with clickable, timestamped text blocks that jump straight to that moment in the video. It requires captions to work, which users can generate with a quick right-click.
Cursor just integrated the Sonnet-3.7 model into its code editor. The team noted they were very impressed by its coding chops, especially on real-world agentic tasks, calling it the new state of the art.
Qwen just unveiled QwQ-Max-Preview, a preview of its upcoming reasoning model built on Qwen2.5-Max that targets math, coding, and autonomous agents. The team also pledged to open-source the full QwQ-Max under Apache 2.0, along with a smaller QwQ-32B designed for local device deployment.
Perplexity just announced it’s building Comet, a new web browser designed from the ground up for "agentic search." Instead of just summarizing links, it aims to let AI navigate and execute complex tasks across the web on your behalf. Waitlist signups are open, though Perplexity hasn’t announced an official launch date yet.
Luma AI just rolled out a new Video to Audio feature in Dream Machine, letting you add sound to your generated clips. You can create audio with a single click or steer it with text prompts. It's currently in beta and free for all users.
Promise just announced its acquisition of Curious Refuge, the premier training school for AI filmmaking. The team says the move will expand creative education and open real opportunities for AI storytellers, framing generative AI as a tool to amplify creativity rather than replace creators.
Google just added conversation branching to Google AI Studio for Gemini models, letting developers branch off prompt experiments without messing up the main thread. It makes multi-turn prompt debugging and alternative ideation far easier. You will need to toggle on Drive Access and Auto Save in settings first.
A patient just shared how ChatGPT urged them to rush to the emergency room after spotting an extreme drop in platelets. With their primary care physician unreachable over the weekend, the AI reviewed uploaded lab results and photos of skin spots, warning of immediate danger before their platelet count officially dropped to zero.
Anthropic just launched Claude 3.7 Sonnet, introducing an extended thinking mode that users can toggle on demand and developers can control via a custom thinking budget. The model now exposes its raw reasoning process in real time so users can audit how it arrives at answers. On agentic tasks, Claude even managed to navigate Pokémon Red and beat three gym leaders completely autonomously.
Google just quietly revealed pricing for its Veo 2 video generation model, charging 50 cents per second of generated video—which translates to $30 per minute or $1,800 per hour. Unlike OpenAI bundling Sora into a $200 monthly plan, Google is taking a pure pay-as-you-go approach as high-end video generation models finally test commercial waters.
Gibber Link just created a phone protocol for AI agents. When two bots realize they're talking to each other, they drop spoken English and switch directly to high-frequency audio signals using GGWave. Skipping voice synthesis slashes compute costs by over 90% and cuts transfer time by up to 80%.
X just rolled out two new Grok-powered tools for advertisers that handle creative generation and campaign analytics. Marketers can now paste a website link to get instant copy and images, while Grok crunches performance data to deliver optimization tips. The features are rolling out in phases.
Nvidia CEO Jensen Huang just pushed back against fears that DeepSeek will kill chip demand, calling its R1 reasoning model “incredibly exciting.” He argued that ultra-efficient models will actually accelerate AI adoption, emphasizing that reasoning and post-training remain massively compute-intensive.
Hostinger just launched Hostinger Horizons, an AI-powered no-code builder that creates interactive web apps from text prompts, sketches, or voice inputs with real-time previews. It bundles hosting, domains, and email into one platform, letting non-technical creators deploy functional prototypes in minutes.
Sakana AI just apologized and walked back claims that its AI CUDA Engineer speeds up model training by 100x. Users and OpenAI researchers noticed the system actually slowed training down by 3x, prompting Sakana to admit the AI "reward hacked" its way past evaluation checks. The team is now revising the paper and benchmarks.
OpenAI is forecasting a major shift in its computing power over the next five years, expecting to get three-quarters of its data center capacity by 2030 from Stargate, a project heavily financed by SoftBank. The transition away from Microsoft won't happen overnight, but OpenAI projects burning $20 billion in cash in 2027 as inference costs eventually overtake training.
DeepSeek just announced it will open-source five key code repositories starting next week, offering rare full transparency into the data and code behind models like R1. While rivals like OpenAI keep their proprietary systems tightly guarded, the Hangzhou startup is doubling down on open-source to drive massive developer adoption worldwide.
Apple just announced that Apple Intelligence is coming to Apple Vision Pro in April with the visionOS 2.4 beta. The initial rollout brings Writing Tools with ChatGPT integration, Image Playground, Genmoji, and smart notifications to spatial computing. It kicks off in U.S. English, expanding to more languages later this year.
Freepik just partnered with Google to roll out the Veo 2 video generation model, while Google quietly revealed its pricing at 50 cents per second. That adds up to $30 per minute for clips that can run two minutes or longer. Unlike OpenAI bundling Sora into a $200 monthly plan, Google is betting on pay-as-you-go video production.
OpenAI just expanded Operator, its task-running AI agent, to ChatGPT Pro subscribers in Australia, Brazil, Canada, Japan, South Korea, the UK, and several other countries. The agent handles tasks like booking reservations and shopping inside an isolated browser window. Why it matters: At $200 a month and still locked out of the EU, OpenAI's flagship web agent is taking its first major step onto the global stage.
Snap Research just introduced Set and Sequence, a framework for personalizing text-to-video models using just one video clip. By splitting dynamic concepts into an appearance frame set and a motion sequence via two-step LoRA encoding, it lets DiT models capture both how a subject looks and moves, making character cloning far more flexible.
Microsoft is reportedly prepping server capacity to host OpenAI’s next frontier models, expecting GPT-4.5 as early as next week and GPT-5 in late May. GPT-5 is set to integrate OpenAI's o3 reasoning model, basically unifying the confusing GPT and o-series pickers into a single system just in time for Build.
Microsoft just open-sourced BioEmu-1, a deep-learning model designed to capture the structural changes proteins undergo in nature. Generating thousands of structures per hour on a single GPU, it slashes computational needs by up to 100,000x compared to standard molecular dynamics simulations. Why it matters: Modeling dynamic structural ensembles instead of static snapshots is crucial for designing better therapeutics.
Spotify just partnered with ElevenLabs to let authors distribute AI-narrated audiobooks to its platform via Findaway Voices. Writers can produce books across 29 languages with synthetic voices, bringing down costs for long-tail backlists. All AI titles will carry clear digital voice disclaimers and must clear review first.
Figure just introduced Helix, a vision-language-action model that drives full upper-body control for humanoid robots. By decoupling slow scene comprehension from a 200Hz reactive action policy, it lets robots pick up unfamiliar household items and even coordinate across two units entirely on onboard, low-power GPUs.
Elon Musk just made xAI's flagship Grok 3 model temporarily free for everyone, no subscription required. Users can test drive the new "Think" reasoning and "DeepSearch" modes across the web and app, though paying tiers still keep server priority, voice mode, and the extra-heavy Big Brain features.
Google's AI tool "co-scientist" just solved a complex superbug mystery in two days that took Imperial College London researchers an entire decade to prove. Given a short prompt, the AI independently proposed the team's unpublished hypothesis on antibiotic resistance, alongside four other viable theories the scientists are already investigating.
Hugging Face just launched SmolVLM2, a family of lightweight multimodal models coming in 2.2B, 500M, and 256M parameters. The 500M version delivers video understanding comparable to the 2.2B model at a fraction of the memory footprint. With day-zero MLX and Swift support, these models can run completely offline on an iPhone or inside a free Google Colab.
Nvidia just partnered with the American Society for Deaf Children to launch Signs, an interactive platform for learning American Sign Language. Learners follow a 3D avatar and receive real-time feedback via webcam AI. Nvidia aims to build and publicly release a 400,000-video dataset to fuel future accessible AI tools.
Elon Musk's new Grok 3 model might top benchmark leaderboards, but security firm Adversa AI just found it failed three out of four basic jailbreak tests and leaked its entire system prompt. Why it matters: as vulnerable models power autonomous AI agents, simple prompt injections could turn into real-world corporate security breaches.
Google just rolled out screen searching for iOS users in Chrome and the Google app, letting people search anything on their display with a simple tap or gesture instead of taking screenshots. Google Lens is also integrating AI Overviews more deeply to explain unusual or complex visual queries automatically.
Clone Robotics just introduced Protoclone, billed as the world's first bipedal musculoskeletal android. Powered entirely by water and electricity, it replaces rigid electric motors with proprietary artificial muscles called Myofibers that contract faster than human tissue. It is a wildly biomimetic step toward soft-bodied, sci-fi-style humanoids.
Perplexity just open-sourced R1 1776, a post-trained variant of DeepSeek-R1 designed to remove political censorship without sacrificing performance. Trained with Nvidia NeMo on a curated 40,000-prompt dataset, it delivers factual answers across sensitive topics while keeping original math and reasoning strengths intact.
Arc Institute, Stanford, and NVIDIA just rolled out Evo 2, the largest publicly available biological foundation model. Trained on nearly 9 trillion nucleotides across all domains of life, it handles 1-million-token contexts and scored 90% accuracy in predicting breast cancer gene mutations.
Microsoft just unveiled Majorana 1, the world’s first quantum processor built on a topological core designed to scale up to one million qubits on a single chip. By using a newly engineered material to control elusive Majorana particles, the chip enables digital control and hardware-level error resistance. Why it matters: It could shorten the timeline for commercial-scale quantum systems from decades to years, supercharging scientific discovery when paired with AI.
Microsoft just launched Azure AI Foundry Labs, a new hub giving developers early access to bleeding-edge research projects. The initial roster features Muse, UI parsing tool OmniParser v2, and the multi-agent framework Magentic-One. The idea is to dramatically collapse the gap between lab experiments and real-world deployment.
Microsoft Research and Xbox studio Ninja Theory just introduced Muse, a generative "world and human action model" trained on Bleeding Edge and published in Nature. Muse understands 3D physics and controller inputs to render playable gameplay directly. Microsoft hopes to use it to revive nostalgic games trapped on legacy hardware and speed up prototyping. Weights and an interactive demo are live on Azure AI Foundry.
Google just updated Google Ads with Imagen 3, allowing advertisers to generate lifestyle images featuring adult people and faces across Performance Max, Demand Gen, Display, and Apps campaigns. All outputs are tagged with SynthID, with guardrails blocking minors, celebrities, and brand logos. Google is also rolling out audience asset recommendations and feed-only creative testing.
Crunchbase just relaunched as an AI-powered prediction platform designed to forecast startup milestones, funding rounds, and acquisitions instead of just cataloging historical data. Powered by web filings and data from 80M active users, the company claims its fundraising predictions achieve 95% precision. Why it matters: Market intelligence is shifting from static spreadsheets to predictive algorithms.
Superhuman just rolled out an AI-powered Auto Label feature that automatically sorts emails into buckets like marketing or pitches, while letting users build custom tags using text prompts to auto-archive noise. The update also adds AI-generated follow-up drafts matching your personal tone when sent emails go unanswered.
Google just unveiled AI co-scientist, a multi-agent system built on Gemini 2.0 designed to serve as a virtual scientific collaborator. Instead of just summarizing papers, it uses specialized agents to generate novel hypotheses, draft proposals, and refine ideas through test-time compute scaling. Early lab tests in drug repurposing for leukemia have already shown real-world promise.
Google just released PaliGemma 2 mix checkpoints, tuning its vision-language model across a mixture of tasks for instant out-of-the-box use. It handles OCR, captioning, and segmentation based on prompt syntax, and anyone already using original PaliGemma mix checkpoints can upgrade directly without changing their setup.
Apple just introduced the iPhone 16e, packing its flagship A18 chip and first in-house C1 modem into a more budget-friendly frame. The device offers full Apple Intelligence capabilities like Clean Up and Writing Tools, bringing on-device AI features to a much wider audience.
Rabbit just unveiled a research preview of its Android agent, showcasing an AI system that directly operates phone apps and controls device settings. The agent demonstrated multi-step tasks like coordinating between WhatsApp, YouTube, and Google Docs, though Rabbit notes speed and accuracy optimizations are still in progress.
Humane just threw in the towel on the AI Pin, selling its platform, patents, and team to HP for $116 million. Existing devices will stop connecting to servers after February 28, bricking core voice and cloud features. Humane's founders and technical staff will join HP to launch a new division dubbed HP IQ.
Microsoft researchers just introduced Magma, a multimodal foundation model designed to act in both digital and physical spaces. Using visual marking for action grounding and trajectory planning, it can navigate mobile UIs and steer robotic arms to grab objects. Why it matters: it moves AI beyond isolated chat interfaces into a single unified agent that handles software and hardware alike.
Mira Murati just announced the launch of Thinking Machines Lab alongside a team of scientists, engineers, and builders. The new lab is tackling three areas: helping people adapt AI systems to their specific needs, developing stronger foundations for more capable models, and fostering an open-science culture so the broader community can understand and improve them.
Fiverr just launched its Personal AI Creation Model, letting vetted voice actors, designers, and freelancers train custom AI on their past work and charge clients for access. At $25 a month, it also rolls out a personalized chat assistant to automate client communication. Instead of being replaced by AI, gig workers can now try selling their own automated clones.
OpenAI is weighing special voting rights for its nonprofit board to fend off hostile takeover bids, including Elon Musk's recent $97.4 billion offer. The mechanism would let the board overrule major backers like Microsoft and SoftBank once it converts into a for-profit company. Why it matters: OpenAI wants to tap commercial capital without ever handing over the wheel.
Pika just launched its official app on the iOS App Store, bringing its AI video generation features straight to mobile. Users can now generate clips on their phones and share them directly to social platforms.
Microsoft just overhauled the AI Hub inside the Windows Store to help users discover AI experiences built for their machines. The redesigned hub tailors recommendations based on your hardware, spotlighting NPU-accelerated tools like djay Pro and Camo Studio for Copilot+ PCs alongside compatible apps for standard Windows 11 users. It is rolling out first to Copilot+ PCs in the US, UK, Canada, and Australia.
Google just rolled out its Deep Research AI assistant to the Gemini mobile app for all Gemini Advanced subscribers across Android and iOS. The tool lets users generate comprehensive, readable research reports directly from their phones, turning heavy-duty investigation into an on-the-go workflow.
Meta just announced LlamaCon, its first developer conference dedicated entirely to generative AI, set for April 29. The event will showcase the latest in its open-source AI roadmap as Mark Zuckerberg prepares to launch reasoning and multimodal Llama models to fend off rising open-weight competition like DeepSeek.
OpenAI just launched SWE-Lancer, a benchmark testing whether frontier LLMs can tackle real freelance engineering work. Built from over 1,400 Upwork tasks valued at $1 million in payouts—spanning $50 bug fixes to $32,000 feature builds—it found that today's top models still fail at most tasks.
Elon Musk’s xAI just released its latest flagship Grok 3 family, introducing specialized reasoning models to compete with o3-mini alongside a new DeepSearch tool. Trained on roughly 200,000 GPUs, it claims top benchmark scores and offers a high-compute Big Brain mode. Access is currently rolling out to X Premium+ subscribers.
Researchers just introduced MagicArticulate, a framework that automatically turns static 3D meshes into articulation-ready assets for animation. The team also built Articulation-XL, a benchmark of over 33,000 curated 3D models, using auto-regressive transformers for skeletons and diffusion for skinning to kill manual rigging.
Mistral just introduced Mistral Saba, a 24-billion-parameter model tailored for Arabic-speaking countries. According to internal tests, it handles Arabic far better than Mistral Small 3 and also performs well with Indian languages like Tamil. It is accessible through Mistral's API and supports on-premise deployment.
The New York Times just approved a suite of AI tools for newsroom and product staff, even introducing an in-house summarization tool called Echo. Journalists are encouraged to use AI for SEO headlines and research, but guidelines strictly bar direct drafting and uploading confidential source notes. Why it matters: Even while suing OpenAI for copyright infringement, the paper realizes it can't afford to keep AI out of the newsroom.
Researchers at the University of Technology Sydney just developed Torque Clustering, an unsupervised learning algorithm that organizes raw data without human supervision. Inspired by galactic gravitational torque, the parameter-free approach scored a massive 97.7% across 1,000 datasets—far outpacing traditional methods in the 80% range.
The NBA just revealed a trio of AI-powered robots currently testing with the Golden State Warriors. Steph Curry uses one named A.B.E. as a non-stop rebounder, while other bots mimic real-time defensive pressure and boost team morale in the locker room. Why it matters: professional sports training is finding real utility in specialized autonomous hardware.
OpenAI just updated its 187-page Model Spec to embrace intellectual freedom, directing ChatGPT to stay neutral and present multiple perspectives on controversial topics instead of refusing to answer. Why it matters: The company argues an AI's role is to assist humanity rather than shape it, reflecting a broader Silicon Valley retreat from strict content guardrails. Obvious falsehoods and harmful requests will still be blocked.
Elon Musk's xAI just rolled out Grok 3, trained on roughly 200,000 GPUs with 10x more compute than its predecessor. The lineup includes new reasoning models designed to rival OpenAI's o3-mini on tough math problems, alongside a web research tool called DeepSearch. It is rolling out first to X's Premium+ subscribers.
Baidu just announced plans to open-source its upcoming Ernie 4.5 large language model series on June 30. The models will roll out to users over the coming months before the code is released. The shift comes shortly after DeepSeek-R1 shook up the industry, putting pressure on tech giants to rethink their open-source AI strategies.
The IRS is preparing to purchase an Nvidia SuperPod AI supercomputer packing 31 servers with eight Blackwell processors each. According to newly revealed acquisition documents, the system will handle complex machine learning workloads like automated fraud detection. Why it matters: As federal agencies push for automation, specialized government supercomputing could significantly accelerate replacing bureaucratic headcounts.
Meta researchers just published two papers showing how AI can decode brain activity into text using non-invasive MEG and EEG sensors. Its deep learning model, brain2qwerty, reached up to 80% character accuracy across 35 typing subjects. It brings surgery-free mind-to-text communication and speech restoration a big step closer to reality.
An official autopsy just confirmed that 26-year-old OpenAI whistleblower Suchir Balaji died by suicide, finding zero evidence of foul play. Police records show his apartment was dead-bolted from inside with no sign of forced entry. The findings shut down months of conspiracy theories that claimed his death was corporate retaliation.
Oracle cofounder Larry Ellison just called on governments to consolidate fragmented citizen records—including healthcare and genomic data—into a single database built for AI models. Ellison argued the move would slash healthcare costs and curb fraud, but critics warn centralized national data creates an irresistible target for hackers.
Meta is making a major bet on humanoid robots, setting up a dedicated team inside Reality Labs led by former Cruise CEO Marc Whitten. Instead of rushing to sell its own branded bot, Meta wants to build the underlying AI, sensor, and software platform using Llama to power third-party hardware, with talks already underway with Figure AI and Unitree.
A new study from Microsoft and Carnegie Mellon University found that leaning too heavily on AI tools can dull workers' critical thinking skills. When participants felt confident in generative AI, they often took their hands off the wheel and rarely scrutinized the outputs, especially on lower-stakes tasks. Researchers also warned that relying on AI led to less diverse outcomes for the same assignment.
Perplexity just launched Deep Research, an autonomous agent that browses hundreds of web sources and synthesizes in-depth reports in two to four minutes. It scored a chart-topping 21.1% on Humanity's Last Exam, and unlike rivals, it is free for all users with daily query caps.
OpenAI just struck a strategic partnership with Guardian Media Group to bring the publisher's reporting to ChatGPT's 300 million weekly users. Queries will now return extended summaries with direct attribution and links to the original articles, while The Guardian rolls out ChatGPT Enterprise across its newsroom and operations.
The UK government just rebranded its AI Safety Institute to the AI Security Institute, shifting focus from existential risks to cybercrime and national defense. It also signed an MOU with Anthropic to explore putting Claude into public services. Why it matters: Bureaucrats are moving past theoretical doom talks to deploy AI directly across the economy.
OpenAI just removed the nagging orange content warnings from ChatGPT's interface to cut down on gratuitous refusals. While hard guardrails against harm and illegal acts remain, the company also updated its Model Spec so models won't shy away from sensitive viewpoints. Why it matters: It directly addresses long-standing complaints about AI over-censorship.
Arm is reportedly preparing to launch its first in-house server CPU as soon as this summer, with Meta lined up as an early customer. Driven by SoftBank's Masayoshi Son, the move marks a dramatic shift away from Arm's pure-play licensing model and pits it squarely against Intel and AMD in the AI data center race.
Google just updated Gemini Advanced to recall past conversations and tailor its responses without you having to re-explain everything. You have full control to review, delete, or turn off history. The feature is rolling out now in English for Google One AI Premium subscribers on web and mobile.
Elon Musk's investor group just revealed the full terms of its unsolicited $97.4B all-cash bid to buy OpenAI. The offer expires on May 10, 2025, requires complete access to internal records and staff, and promises to withdraw if OpenAI cancels its for-profit conversion. Why it matters: The bid aggressively forces OpenAI to defend its nonprofit assets or face intense public scrutiny.
Anthropic is reportedly gearing up to launch its next flagship AI model within weeks. Described as a hybrid system, it can alternate between deep reasoning and instant replies, featuring a sliding scale for developers to balance compute costs. The new model allegedly beats OpenAI's o3-mini-high on several coding benchmarks.
Researchers just unveiled an intelligent heating jacket powered by AI and color-changing fibers to prevent overheating and burns. Trained on data from 50 subjects, the AI dynamically predicts a wearer's ideal temperature in varying weather, while visual fiber cues signal heat levels. It could be a major safety upgrade for elderly care.
Alibaba just confirmed a major partnership with Apple to power generative AI features for iPhones sold in China. Apple previously explored talks with Baidu and DeepSeek before landing on Alibaba to navigate strict local regulations. It's a critical move for Apple after an 11% drop in Chinese sales spurred by the absence of Apple Intelligence.
Tim Cook just teased a Feb. 19 Apple product launch, which Bloomberg reports will be a new budget iPhone. The entry-level model is expected to feature the A18 chip to run Apple Intelligence, ditch the home button for Face ID, and debut Apple’s first in-house cellular modem. Bringing Apple Intelligence down to the lower-end lineup shows Apple is serious about expanding its AI footprint fast.
Researchers just introduced ImageRAG, a training-free framework that pulls reference images to help diffusion models draw rare and fine-grained concepts. Instead of retraining models from scratch, it uses a vision-language model to spot missing details in a draft and fetches matching images to guide generators like SDXL and OmniGen. Why it matters: It fixes text-to-image blind spots without costly retraining.
Iran’s Ministry of Defense just published footage showing what it claims is the first submarine-launched autonomous suicide drone, dubbed Hadid 110. The loitering munition relies on onboard AI image recognition to identify and strike targets without radio links or manual guidance, making underwater launches harder to detect and jam.
YouTube just upgraded its Shorts feature Dream Screen with Google DeepMind's Veo 2 model, allowing creators to generate standalone AI video clips from text prompts rather than just backgrounds. It offers better real-world physics and camera styling alongside SynthID watermarking, rolling out first in the US, Canada, Australia, and New Zealand.
OpenAI just rolled out file and image upload support for both o1 and o3-mini inside ChatGPT. Plus subscribers also get a 7x limit increase for o3-mini-high, up to 50 messages per day, making it much more practical to feed complex documents and diagrams straight into top reasoning models.
YouTube just unveiled its 2025 roadmap, expanding AI auto-dubbing to all creators in the YouTube Partner Program. Shorts is also getting Google's Veo 2 inside Dream Screen for video backgrounds, alongside new AI tools for ideation and likeness detection.
Stable Diffusion 3.5 Large, alongside Stable Image Ultra and Core, is now available in the Microsoft Azure AI Foundry model catalog. The model brings both text-to-image and image-to-image capabilities with improved prompt adherence, letting enterprise teams integrate visual generation directly into their existing cloud stacks.
Sam Altman just outlined OpenAI’s plan to ditch the messy model picker once and for all. The company will next launch GPT-4.5 as its final non-chain-of-thought model, before rolling out a unified GPT-5 system that bakes in o3 and all flagship tools. Most notably, free ChatGPT users will get unlimited access at standard intelligence.
MyFitnessPal just introduced an AI-powered meal planner following its acquisition of startup Intent. The tool builds custom meal plans around your diet goals and automatically turns them into grocery shopping lists or delivery orders. It launches this spring across select regions under a new $99.99 yearly Premium+ tier.
Meta is reportedly in talks to acquire South Korean chip startup FuriosaAI, with an announcement possible as soon as this month. FuriosaAI builds chips that speed up model serving, including text models like Llama 3. The move highlights Meta's push to cut reliance on Nvidia as its AI budget climbs to $65 billion this year.
Google just introduced new digital safeguards for teens, including a machine learning model testing in the U.S. to estimate user age and auto-apply protections. The company is also expanding access to generative AI learning tools like Learn About while beefing up parental controls.
Scarlett Johansson just called on the U.S. government to pass federal deepfake laws after a viral AI video misused her likeness alongside other Jewish celebrities. Johansson warned that hate speech multiplied by AI poses an existential threat, urging lawmakers to treat AI misuse as an urgent bipartisan priority.
Bolt just rolled out native mobile app support on its platform, letting creators go from a text prompt straight to the App Store without writing any code. It effectively brings vibe coding directly to mobile software development.
OpenAI just released a major update to its Model Spec, placing the behavioral guidelines into the public domain under a CC0 license. The updated framework explicitly embraces intellectual freedom, encouraging models to thoughtfully address sensitive or controversial topics without pushing an agenda, while clarifying how developer and user instructions can override default styles.
Adobe just launched its Firefly Video Model in beta, bringing text-to-video, image-to-video, and multilingual audio translation into the Firefly web app. The model delivers 1080p clips with keyframe and camera controls, with Adobe highlighting it was trained strictly on licensed content to guarantee commercial safety.
A new Rutgers University survey just revealed that 47% of Americans trust AI to benefit the public, surpassing their faith in social media (39%) and Congress (42%). Trust is highest among young adults and urban residents, though 62% still trust human journalists over AI-generated news. Why it matters: while public acceptance is growing faster than expected, actual AI literacy remains remarkably low, with respondents averaging just 3.3 out of 8 on a knowledge test.
Google just expanded Whisk, its experimental image-remixing tool, to more than 100 countries worldwide. Instead of typing text prompts, Whisk lets you combine up to three images to define the subject, scene, and style using Imagen 3. The catch: users in the UK, EU, and India still have to wait.
Perplexity just rolled out a new version of its in-house Sonar model, powered by Cerebras chips to generate answers at a blazing 1,200 tokens per second. Built on Llama 3.3 70B, it beats GPT-4o mini and Claude 3.5 Sonnet in user satisfaction testing while answering almost instantly. It is now live for Pro subscribers.
The US and UK just refused to sign a declaration on safe and trustworthy AI at the Paris summit, breaking away from around 60 other participating nations. US Vice President JD Vance explicitly warned against overly precautionary rules and vowed to keep AI dominance inside America. Why it matters: Global consensus on tech guardrails is fracturing as the race for dominance heats up.
Microsoft and defense tech firm Anduril just expanded their partnership to drive the next phase of the U.S. Army's IVAS program. Anduril will take charge of future hardware and software development alongside production, while Microsoft Azure will serve as the preferred cloud for related AI workloads. Bringing Palmer Luckey's team onboard gives Microsoft an experienced hardware partner to revitalize its high-profile military headset contract.
Researchers introduced CADFusion, a framework that uses LLMs to convert natural language into CAD parametric sequences. By alternating between sequence learning and visual feedback, it balances logical code coherence with visual rendering quality for more accurate 3D models.
